From pointwise to one common index
Suppose continuous real-valued functions on a compact set converge pointwise to a continuous function . Pointwise convergence lets the required index depend on the point; uniform convergence requires one index for all points. Even on a compact set, continuity alone does not bridge that gap. A monotonicity assumption does.
Definition 16.1.1
(Uniform convergence)
A sequence converges to uniformly on if for every there is one index such that for all and all . Pointwise convergence only asks for an index that may depend on .
Theorem 16.1.2
(Dini's theorem)
Let be compact and nonempty, let be continuous, and suppose pointwise with for every and . Then uniformly on .
The analogous increasing case follows by replacing each function by its negative. The empty set is vacuous, so we assume nonempty.
Proof by compactness
Proof
(Nested open sets and a finite subcover)
Let . Each is continuous, decreases pointwise, and tends to zero. In particular : a decreasing real sequence cannot fall below its limit. Fix and set
Each is open relative to . Pointwise convergence says these sets cover . Monotonicity makes them increasing: . Compactness gives a finite subcover; let be the largest index in it. Since the sets are nested, . Thus for every , and the same is true for whenever . This is precisely uniform convergence.
Proof X-ray
Which hypothesis does which job
The proof is a chain of four steps, each fed by exactly one hypothesis. Continuity of and makes continuous, so the strict sublevel set is open in . Pointwise convergence gives, for each , some with , so the cover . Compactness extracts finitely many that still cover. Monotonicity then collapses the finite cover to a single set: because , the one with the largest index contains all the others, so .
The non-obvious move is the last one. Without monotonicity a finite subcover gives, for each point, some index that works, but no single index works for the whole cover, and the index would not persist for later . The proof closes because also guarantees for every , not only for . Each missing hypothesis is tested by a concrete example below.
Worked examples
Worked example 16.1.3
(A decreasing sequence with an explicit uniform error)
On , let for . Each is continuous, , and pointwise. The limit is continuous, so Dini's theorem applies. Here there is also a direct estimate: , with equality at . Thus the uniform error is exactly . The theorem guarantees convergence but does not itself supply this particular error bound.
Worked example 16.1.4
(Zero at both endpoints)
On , let . Each is continuous and . For , , and , so pointwise. Dini's theorem gives uniform convergence. A direct check: has its maximum at , where it equals .
Worked example 16.1.5
(A nonzero limit)
On , let . These continuous functions decrease in to the continuous function . Here is the constant , so as soon as , and the finite subcover can be taken to be a single set. The example shows that the proof is applied to , not to itself.
Three hypotheses that matter
Counterexample mode
A moving peak needs no common index
Claim: continuous functions on that converge pointwise to converge uniformly. Let on . Every is continuous, and for each fixed : at zero every value is zero; for fixed , eventually and . But , so for every and the convergence is not uniform. The sequence is not pointwise monotone: at , and . Monotonicity in is exactly what rules this example out.
Compactness matters as well, and it is the next hypothesis to test.
Counterexample mode
Without compactness the same sequence fails
Claim: Dini's theorem holds on . The functions are continuous and decrease in to the continuous zero function at each fixed . But at we have , so (in fact it equals ) for every , and the convergence is not uniform. The finite-subcover step is where the proof breaks: the sets cover but no finite subfamily does.
Common mistake
(The limit must be continuous)
The functions decrease on to for and to at , a discontinuous limit, and the convergence is not uniform. For example, satisfies by Bernoulli's inequality, while the limit at is zero. Dini's theorem proves uniform convergence only to a continuous limit; is not open when is discontinuous.
Common mistake
(Monotone in x is not monotone in n)
Dini's monotonicity condition concerns the index: at each fixed . The functions are increasing in but decreasing in ; the second property is the one the theorem uses. Functions that are each monotone in need not be monotone in at all: the tents above are increasing in on , yet .
Exercises and solutions
Quick check
(On , let . Does Dini's theorem apply, and what is the largest error?)
Check continuity, monotonicity in and the limit.
Solution · Answer
Yes: the sequence decreases to the continuous zero function and the maximum is .
Quick check
(If continuous increase pointwise to continuous on compact , explain how the proved version applies.)
Change signs.
Solution · Answer
decreases to , and the absolute errors are unchanged.
Quick check
(On , find a point where .)
Solve for in terms of .
Solution · Answer
. This gives error at least for every , already disproving uniform convergence.
Quick check
(Why can the compactness proof use the largest of finitely many indices?)
Look at how the sets are ordered.
Solution · Answer
The sublevel sets are nested upward, so the set with the largest index contains every set in the finite subcover.